Anthropic says these topics are too dangerous to let its Fable 5 model talk about
Ars Technica AI · Kyle Orland · 2026-06-09
Anthropic launches Claude Fable 5, a new Mythos-class model that routes queries about cybersecurity, biology, and chemistry to an older Opus 4.8 model to prevent malicious actors from gaining dangerous capabilities.
Appears in
Extraction
Topics: ai-safetyllm-releasecontent-moderationdual-use-aicybersecurity-ai
Claims
- Fable 5 is Anthropic's first Mythos-class model and surpasses previous Opus frontier models in overall capabilities.
- Fable 5 and Mythos 5 share the same underlying model, but Mythos 5 access is restricted to vetted cyberdefenders through Project Glasswing.
- Sensitive queries on cybersecurity, biology, and chemistry are silently routed to the older Claude Opus 4.8 model rather than answered by Fable 5.
- The safeguards are deliberately tuned stricter than ideal, triggering false positives in under 5% of sessions in testing.
- Fable 5 showed a particularly large benchmark improvement in cybersecurity tasks relative to prior models.
Key quotes
Anthropic said it has tuned these safeguards to be 'stricter than ideal,' meaning the system may occasionally refuse 'harmless requests' in a way that it acknowledges may be frustrating for regular users.
Mythos could give malicious actors assistance in 'causing serious harm that they couldn't have received from other sources.'